# Solver integration result

The actual AIFlow current and thermal commands both accept the fresh candidate
report in `revision-b/solver-integration-reviewed`. This is an isolated integration
test, not the main engineering run and not a production qualification.

## What was exercised

- Pinned local executable: `tools/fields-candidate/bin/adom-fields`, SHA256
  `9bcf9848d1d80739862b6f3b7eff73a1e4252eed976c98214f722fff6350afc4`.
- Same unchanged PCB on the host and in the workspace, SHA256
  `e15c130e125a566dd4c2660585a2eee355b08fea71d6dfee8c17a6c84d40a57c`.
- Explicit current terminals, two-layer stack, exposed-pad heat locations and
  global 5 C adapter from the previously evaluated diagnostic specification.
- Fresh solve at 0.2 mm mesh; no cached report hashes were changed.
- Real `adom-aiflow analyze current --fields ...` and `analyze thermal --fields ...`,
  with the existing route plan and a copy of the saved zone-state measurement.
- Checks preserve every original load current, hot-part dissipation and per-part
  temperature-rise limit. Both commands returned exit 0 and recorded pass=true.

Run `python3 design/run_solver_gate_check.py` to repeat the consumer checks.
Machine-readable results: `solver-integration-review.json`; CLI logs:
`integration-reviewed-current.log` and `integration-reviewed-thermal.log`.

## Additional blocker found and resolved in the integration candidate

The first real consumer check passed current but rejected thermal, because the
original spec still requested six U3.11 vias. The previous workflow had explicitly
acknowledged retaining the layout, but that acknowledgement did not update the
Fields input. This was a workflow inconsistency, not merely a display warning.

`design/fields-integration-spec.json` explicitly records one retained via within
3 mm of U3.11, consistent with `via-count-review.json`. Three are near any U3 ground
pad under AIFlow's different definition. The generic six-via floor was an earlier
assistant/tool target, not a user-specified requirement. Its prior acknowledgement
is carried forward transparently in `thermalViaReview`; no holes were added, no
heat load reduced, and the 5 C thermal limit remains. A second fresh solve with
this exact specification passes the real thermal consumer check.

## Remaining acceptance dependency

Live registry check on 2026-10-14 still lists Fields 0.7.15 as latest. Correction
PR https://wiki.adom.inc/adom/adom-fields/prs/9 remains open; its body requests
review/merge of the dependency stack before shared publication/installation.
See `solver-review-status.json`. The local candidate also has subsequent changes
listed in `tools/fields-candidate/provenance.json`; approval of PR9 alone does not
prove that entire candidate is reviewed.

No shared executable was replaced. The main `design/spec.json` and engineering
run were not promoted or marked passed using experimental evidence. The next
formal integration must select an accepted build covering the required fixes,
adopt the documented input changes, solve again, and run the main current and
thermal commands before proceeding through Fields, nets, silkscreen, 3D, finish
and export. Previously prepared manufacturing files remain prototype artifacts.

This work demonstrates software integration under stated modelling assumptions.
It does not establish enclosure cooling, RF performance, HX711 accuracy, battery
life or rotating-assembly qualification. Firmware remains deferred by the user.